Domain Generalization (DG) requires models to generalize across unseen data distributions. Kernel-based theory reveals a No-Free-Lunch problem: any model with a fixed representation is fundamentally sub-optimal for all possible shifts. While large ensembles mitigate this, they are computationally expensive and remain static once trained, inheriting the same theoretical limitation. We introduce MoE² (Mixture-of-Mixtures of Experts), a framework that uses a single frozen backbone to dynamically synthesize a bespoke adapter for each input, allowing it to continuously adapt its effective kernel. We provide a theoretical grounding for this process, proving our routing mechanism is a principled non-parametric estimator for the optimal Bayes mixture of experts. We derive a generalization bound that cleanly separates the router's estimation error from the reduction in a kernel-mismatch penalty achieved via synthesis. MoE² matches or exceeds state-of-the-art ensemble baselines on major DG benchmarks while using only a single, compact model. MoE² thus provides a theoretically-grounded and lightweight alternative to large-scale ensembles for robust domain generalization.

One-stream transformer-based trackers have shown remarkable success in single object tracking by jointly performing feature extraction and relation modeling. However, the update of temporal context, often propagated via temporal tokens, typically relies on general self-attention mechanisms within the transformer backbone. This paper introduces STTATrack, a novel framework that enhances one-stream tracking by explicitly leveraging the immediate spatial certainty from the current frame’s prediction score map to refine these propagated temporal query tokens. Our core contribution, the Score Temporal Token Attention (STTA) module, generates an embedding from the score map and employs a dual attention mechanism to facilitate bidirectional information flow between this spatial certainty embedding and the existing temporal tokens. This targeted refinement allows temporal tokens to be dynamically adapted based on the most current and spatially precise evidence, leading to improved adaptability and temporal consistency. STTATrack builds upon the ODTrack architecture and demonstrates significant performance improvements on the challenging GOT10k, OTB and UAV123 benchmarks, underscoring the efficacy of explicitly integrating current-frame spatial certainty into the temporal refinement loop.

In application (app) development, effectively harnessing user feedback is crucial for enhancing app quality and user feedback. However, the vast and unstructured nature of user reviews often complicates these efforts, posing challenges in accurately capturing and integrating this feedback into the development processes. We automate the classification of issues in app reviews and examine how these issues correlate with code quality metrics (code smells and bug reports) and development activities (additions, deletions, and time to merge in pull requests). We aim to provide evidence-based guidance for effectively prioritizing and addressing user feedback. Employing a Mining Software Repositories (MSR) approach, we gathered and analyzed reviews from seven open-source Android apps. We evaluated the efficacy of three machine learning models-Support Vector Machines (SVM), BERT, and a fine-tuned GPT-3.5-for classifying issues in app reviews. The GPT-3.5 model achieved the highest accuracy at 95.0%. We found statistically significant correlations between the classified issues, code quality metrics, and development activities. However, these relationships varied across applications, highlighting the complex relationship between user feedback and the development process. Our study highlights the effectiveness of automated tools in identifying and classifying feedback within app reviews. Our automated approach enhances developers' ability to manage feedback effectively and supports optimal resource allocation to improve app quality and user feedback.

Due to the large volume of data and its unstructured nature, managing user feedback via application (app) reviews is a significant challenge for Android developers. This study presents a dashboard to streamline this process using advanced machine learning and analysis techniques. The dashboard employs a fine-tuned Generative Pretrained Transformer (GPT-3.5) model to detect and categorize issues in user reviews automatically. Additional dashboard features include sentiment and toxicity analysis to provide insights into user emotions, potentially negative feedback, and code analysis to identify code smells across different app versions. We conducted a pilot study to evaluate the usability and effectiveness of the dashboard. The results indicate that the dashboard is user-friendly and effective in helping developers manage user feedback and monitor code quality. However, certain limitations were identified, such as dependency on the quality of training data and potential inaccuracies in sentiment and toxicity analysis. This dashboard aims to aid developers in effectively managing app reviews, prioritizing issues, and maintaining high app quality to improve user satisfaction. Tool URL: https://tdresearchgroup.github.io/Review-Pulseldashboard/ Demo Video: https://youtu.be/cT6su8dqh2g